T
E
L
K
O
M
N
I
K
A
T
elec
o
m
m
un
ica
t
io
n,
Co
m
pu
t
ing
,
E
lect
ro
nics
a
nd
Co
ntr
o
l
Vo
l.
18
,
No
.
6
,
Dec
em
b
er
2
0
2
0
,
p
p
.
3
0
26
~3
0
3
2
I
SS
N:
1
6
9
3
-
6
9
3
0
,
ac
cr
ed
ited
First Gr
ad
e
b
y
Kem
en
r
is
tek
d
i
k
ti,
Dec
r
ee
No
: 2
1
/E/KPT
/2
0
1
8
DOI
: 1
0
.
1
2
9
2
8
/TE
L
KOM
NI
K
A.
v
1
8
i6
.
1
6
1
7
4
3026
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//jo
u
r
n
a
l.u
a
d
.
a
c.
id
/in
d
ex
.
p
h
p
/TELK
OM
N
I
K
A
The
det
ection o
f
ha
ndg
uns from
li
v
e
-
v
ideo
in
real
-
ti
me ba
sed o
n
deep learning
M
o
ha
m
m
ed
G
ha
za
l
,
Na
j
wa
n Wa
is
i
,
Na
wa
l A
bd
ull
a
h
De
p
a
rt
m
e
n
t
o
f
Co
m
p
u
ter E
n
g
in
e
e
rin
g
Tec
h
n
o
lo
g
y
,
No
rt
h
e
rn
Tec
h
n
ica
l
Un
iv
e
rsity
,
M
o
su
l,
Ira
q
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ma
r
2
7
,
2
0
2
0
R
ev
is
ed
J
u
n
1
2
,
2
0
2
0
Acc
ep
ted
J
u
l 9
,
2
0
2
0
M
a
n
y
p
e
o
p
le
h
a
v
e
b
e
e
n
k
il
led
i
n
d
isc
rimin
a
tely
b
y
th
e
u
se
o
f
h
a
n
d
g
u
n
s
i
n
d
iffere
n
t
c
o
u
n
tri
e
s.
Terro
rist
a
c
ts,
o
n
li
n
e
fig
h
ti
n
g
g
a
m
e
s
a
n
d
m
e
n
tall
y
d
istu
r
b
e
d
p
e
o
p
le are
c
o
n
sid
e
re
d
t
h
e
c
o
m
m
o
n
re
a
so
n
s fo
r
th
e
se
c
ri
m
e
s.
A rea
l
-
ti
m
e
h
a
n
d
g
u
n
s
d
e
tec
ti
o
n
su
r
v
e
il
l
a
n
c
e
sy
ste
m
is
b
u
il
t
to
o
v
e
rc
o
m
e
th
e
se
b
a
d
a
c
ts,
b
a
se
d
o
n
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
r
a
l
n
e
two
rk
s
(CNN
s).
T
h
is
m
e
th
o
d
is
fo
c
u
se
d
o
n
t
h
e
d
e
tec
ti
o
n
o
f
d
iffere
n
t
w
e
a
p
o
n
s,
s
u
c
h
a
s
(h
a
n
d
g
u
n
a
n
d
r
ifl
e
s).
Th
e
id
e
n
ti
fica
ti
o
n
o
f
h
a
n
d
g
u
n
s
fro
m
su
rv
e
il
lan
c
e
c
a
m
e
ra
s
a
n
d
ima
g
e
s
re
q
u
ires
m
o
n
it
o
r
in
g
b
y
h
u
m
a
n
su
p
e
rv
is
o
r,
t
h
a
t
c
a
n
c
a
u
se
e
rro
rs.
To
o
v
e
rc
o
m
e
th
is
issu
e
,
th
e
d
e
sig
n
e
d
d
e
tec
ti
o
n
sy
ste
m
se
n
d
s
a
n
a
lert
m
e
ss
a
g
e
t
o
th
e
su
p
e
r
v
i
so
r
wh
e
n
a
we
a
p
o
n
is
d
e
tec
ted
.
In
th
e
p
r
o
p
o
se
d
d
e
tec
ti
o
n
sy
ste
m
,
a
p
re
-
train
e
d
d
e
e
p
lea
rn
in
g
m
o
d
e
l
M
o
b
il
e
Ne
tV3
-
S
S
DLit
e
is
u
se
d
to
p
e
rf
o
rm
th
e
h
a
n
d
g
u
n
d
e
tec
ti
o
n
o
p
e
ra
ti
o
n
.
Th
is
m
o
d
e
l
h
a
s
b
e
e
n
se
lec
ted
b
e
c
a
u
se
it
i
s
fa
st
a
n
d
a
c
c
u
ra
te
in
in
fe
rin
g
t
o
in
teg
ra
te
n
e
two
rk
fo
r
d
e
tec
ti
n
g
a
n
d
c
las
sify
in
g
we
a
p
o
n
s
in
ima
g
e
s.
Th
e
e
x
p
e
rime
n
tal
re
su
lt
u
sin
g
g
lo
b
a
l
h
a
n
d
g
u
n
s
d
a
tas
e
ts
o
f
v
a
rio
u
s
we
a
p
o
n
s
sh
o
we
d
th
a
t
th
e
u
se
o
f
M
o
b
il
e
Ne
tV3
wi
th
S
S
DLi
te
m
o
d
e
l
b
o
t
h
e
n
h
a
n
c
e
th
e
a
c
c
u
ra
c
y
lev
e
l
in
id
e
n
ti
fy
i
n
g
th
e
re
a
l
t
ime
h
a
n
d
g
u
n
s d
e
t
e
c
ti
o
n
.
K
ey
w
o
r
d
s
:
C
NN
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
Dee
p
le
ar
n
in
g
Han
d
g
u
n
Mo
b
ileNetV3
SS
DL
ite
W
ea
p
o
n
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r
th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Mo
h
am
m
ed
T
alal
Gh
az
al,
De
p
a
rt
m
e
n
t
o
f
Co
m
p
u
ter E
n
g
in
e
e
rin
g
Tec
h
n
o
lo
g
y
,
No
rth
e
rn
Tec
h
n
ica
l
Un
iv
e
rsity
,
Mo
s
u
l,
I
r
aq
.
E
m
ail:
m
o
h
am
m
e
d
.
g
h
az
al@
n
t
u
.
ed
u
.
i
q
1.
I
NT
RO
D
UCT
I
O
N
C
r
im
e
r
ates
u
s
in
g
h
an
d
g
u
n
s
h
a
v
e
in
cr
ea
s
ed
in
m
a
n
y
p
lace
s
in
th
e
wo
r
ld
[
1
]
,
esp
ec
ially
in
t
h
e
co
u
n
tr
ies
th
at
ca
n
n
o
t
r
estrict
wea
p
o
n
s
to
th
e
s
tate.
D
aily
cr
im
es
o
cc
u
r
with
v
ar
io
u
s
wea
p
o
n
s
in
p
u
b
li
c
p
lace
s
an
d
p
lace
s
o
f
wo
r
s
h
ip
.
T
h
e
Un
ited
Natio
n
s
Of
f
ice
f
o
r
Dr
u
g
s
an
d
C
r
im
e
(
UNOD
C
)
r
ep
o
r
ted
th
e
s
tati
s
tics
f
o
r
th
e
n
u
m
b
er
o
f
cr
im
es
in
v
o
l
v
in
g
wea
p
o
n
s
p
er
1
0
0
,
0
0
0
h
ab
itan
ts
is
v
er
y
h
ig
h
in
m
a
n
y
co
u
n
tr
ies,
f
o
r
ex
am
p
le.
,
1
0
.
0
8
in
C
en
tr
al
I
r
aq
,
2
.
5
in
L
ib
y
a
a
n
d
2
.
2
0
in
Sy
r
ia
[
2
]
.
I
n
ir
aq
,
p
o
li
ce
tr
ied
to
im
p
o
s
e
th
eir
co
n
tr
o
l
in
th
ese
p
lace
s
b
y
d
ep
lo
y
in
g
s
ec
u
r
it
y
m
e
n
am
o
n
g
th
e
cr
o
wd
s
an
d
,
b
y
in
s
tallin
g
s
u
r
v
eillan
ce
ca
m
er
as
.
T
h
ese
m
eth
o
d
s
ar
e
in
ef
f
ec
tiv
e
an
d
ex
p
e
n
s
iv
e
,
esp
ec
ially
wh
en
m
u
ltip
le
v
id
eo
s
tr
ea
m
s
ar
e
p
r
esen
t
o
n
th
e
c
am
er
a
s
y
s
tem
s
.
T
h
e
s
u
g
g
ested
m
eth
o
d
o
f
h
an
d
g
u
n
d
etec
tio
n
s
y
s
tem
wh
ich
is
b
ased
o
n
th
e
v
id
eo
s
tr
ea
m
s
f
r
o
m
s
u
r
v
eillan
ce
ca
m
er
as
ca
n
aler
t
th
e
s
ec
u
r
ity
m
en
i
f
a
p
er
s
o
n
with
g
u
n
is
d
etec
ted
i
n
r
ea
l
-
tim
e,
wh
ic
h
will
lead
t
o
s
p
ee
d
u
p
h
an
d
g
u
n
d
etec
tio
n
b
ef
o
r
e
th
e
c
r
im
e
o
cc
u
r
e
.
T
h
e
p
r
o
p
o
s
ed
s
y
s
tem
f
o
r
h
an
d
g
u
n
d
etec
tio
n
in
r
ea
l
-
tim
e
is
b
ased
o
n
co
n
v
o
l
u
tio
n
al
n
eu
r
al
n
etwo
r
k
(
C
NN)
.
T
h
e
C
NNs
h
as
b
ee
n
r
ap
id
ly
g
r
o
win
g
in
c
o
m
p
u
ter
v
is
io
n
ar
ea
d
u
r
in
g
th
e
p
ast
f
ew
y
ea
r
s
[
3
-
7
]
.
I
n
co
m
p
u
ter
v
is
io
n
th
er
e
a
r
e
t
h
r
e
e
o
b
ject
d
etec
tio
n
m
o
d
els
h
av
e
b
ee
n
an
aly
ze
d
(
f
aster
r
eg
i
o
n
-
b
ased
co
n
v
o
lu
tio
n
al
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
T
h
e
d
etec
tio
n
o
f
h
a
n
d
g
u
n
s
fr
o
m
live
-
vid
eo
in
r
ea
l time
b
a
s
e
d
o
n
d
ee
p
lea
r
n
in
g
(
Mu
h
a
mme
d
Gh
a
z
a
l
)
3027
n
eu
r
al
n
etwo
r
k
s
(
F
R
-
C
NN)
,
r
eg
io
n
b
ased
-
f
u
lly
co
n
v
o
lu
tio
n
al
n
etwo
r
k
s
(R
-
FC
N)
an
d
s
in
g
le
s
h
o
t
d
e
tecto
r
(
SS
D)
)
[
8
]
.
T
o
en
h
an
ce
th
e
s
p
ee
d
,
ac
cu
r
ac
y
an
d
p
er
f
o
r
m
a
n
c
e,
th
ese
m
o
d
els
h
a
v
e
b
ee
n
co
m
b
in
ed
with
d
if
f
er
en
t
f
ea
tu
r
e
ex
tr
ac
to
r
s
s
u
ch
as
(
VGG,
Mo
b
ilen
et
an
d
R
esn
et
-
101)
,
f
o
r
th
at
r
ea
s
o
n
th
e
Mo
b
ile
Netv
3
with
SS
DL
ite
m
o
d
el
h
av
e
b
ee
n
u
s
ed
in
th
e
p
r
o
p
o
s
ed
s
y
s
tem
[
9
]
.
T
h
e
Mo
b
ileNetv
3
is
%2
7
f
aster
th
an
Mo
b
ileNetv
2
,
wh
ile
m
ain
tain
i
n
g
s
im
ilar
m
AP
.
M
o
r
eo
v
e
r
,
its
clas
s
if
icatio
n
is
b
ased
o
n
p
r
e
-
tr
ain
ed
d
ataset
o
n
th
e
I
m
ag
eNe
t.
A
wid
ely
u
s
ed
d
ataset
is
I
m
ag
e
Net,
wh
ich
is
an
o
b
ject
d
etec
tio
n
d
ataset
wh
ich
co
n
tai
n
es
ab
o
u
t
1
.
2
8
m
illi
o
n
im
ag
es
with
ap
p
r
o
x
im
ately
1
,
0
0
0
o
b
ject
class
es
[
1
0
]
.
Ou
r
ap
p
r
o
ac
h
is
to
u
s
e
a
g
lo
b
al
d
ataset
co
n
tain
in
g
im
ag
es
o
f
th
e
m
o
s
t u
s
ed
ty
p
es
o
f
wea
p
o
n
s
,
to
b
e
clo
s
e
to
r
ea
l c
ir
cu
m
s
tan
ce
,
th
e
h
an
d
g
u
n
s
d
ataset
s
f
o
r
b
o
th
th
e
s
lid
in
g
win
d
o
w
an
d
r
eg
i
o
n
p
r
o
p
o
s
als
ap
p
r
o
ac
h
s
h
av
e
b
ee
n
s
elec
ted
[
1
1
]
,
in
a
d
d
itio
n
to
,
wea
p
o
n
d
a
taset
in
[
1
2
]
.
T
h
ese
co
n
tain
im
ag
es
o
f
wea
p
o
n
s
in
a
v
ar
iety
o
f
d
if
f
er
e
n
t
co
n
tex
ts
,
s
itu
atio
n
s
,
an
d
o
r
ien
tatio
n
s
.
T
h
e
d
atasets
ar
e
tr
ain
ed
o
n
Mo
b
ileNetv
3
_
SS
DL
ite
C
o
n
v
o
lu
tio
n
al
Neu
r
al
Netwo
r
k
C
NN.
2.
RE
L
AT
E
D
WO
RK
S
T
h
e
r
e
a
r
e
t
w
o
m
e
t
h
o
d
s
f
o
r
h
a
n
d
g
u
n
d
e
t
e
c
t
i
o
n
s
y
s
t
e
m
:
t
h
e
f
i
r
s
t
i
s
b
as
e
d
o
n
m
e
t
a
l
d
e
t
e
c
t
i
o
n
u
s
i
n
g
X
-
r
a
y
o
r
m
i
l
l
i
m
et
r
i
c
w
a
v
e
i
m
ag
e
s
,
i
t
u
s
e
d
i
n
a
i
r
p
o
r
ts
t
o
c
h
e
ck
t
h
e
p
a
s
s
e
n
g
e
r
l
u
g
g
a
g
e
’
s
[
1
3
]
.
t
h
e
s
e
m
et
h
o
d
s
a
r
e
u
n
a
b
l
e
t
o
n
o
n
-
m
e
t
a
l
l
i
c
h
a
n
d
g
u
n
d
e
t
e
c
t
i
o
n
a
n
d
i
t
i
s
l
i
m
i
t
e
d
t
o
u
s
e
i
n
s
p
e
c
i
f
i
c
p
l
a
ce
s
.
T
h
e
s
e
c
o
n
d
i
s
b
a
s
e
d
o
n
R
GB
i
m
a
g
e
s
[
1
4
,
1
5
]
,
f
r
o
m
t
h
e
r
e
c
e
n
t
p
a
p
e
r
s
f
a
s
t
r
e
g
i
o
n
-
b
a
s
e
d
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
t
w
o
r
k
(
F
R
C
NN
)
a
n
d
r
e
g
i
o
n
-
b
a
s
e
d
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
tw
o
r
ks
(
R
C
N
N
)
b
as
e
d
m
o
d
e
ls
a
r
e
ap
p
l
i
e
d
t
o
a
u
t
o
m
a
ti
c
h
a
n
d
g
u
n
d
e
t
e
c
t
i
o
n
s
y
s
te
m
i
n
s
u
r
v
e
i
l
l
a
n
ce
v
i
d
e
o
s
[
1
2
]
,
a
n
o
t
h
e
r
a
u
t
h
o
r
u
s
e
a
n
o
v
e
l
o
b
j
e
c
t
d
e
t
e
c
t
i
o
n
a
l
g
o
r
i
t
h
m
s
t
o
d
e
t
e
c
t
t
h
e
v
i
s
u
a
l
k
n
i
f
e
f
o
r
t
h
e
g
i
v
e
n
v
i
d
e
o
d
a
t
a
s
e
t
[
1
6
]
,
t
h
e
f
o
r
e
g
r
o
u
n
d
s
e
g
m
e
n
t
a
t
i
o
n
i
s
u
s
e
d
f
o
r
f
e
a
t
u
r
e
s
e
x
t
r
a
c
ti
o
n
,
t
h
e
f
e
a
t
u
r
e
d
e
t
e
c
t
i
o
n
p
e
r
f
o
r
m
e
d
b
y
F
e
a
t
u
r
es
f
r
o
m
a
cc
e
l
e
r
a
t
e
d
s
e
g
m
e
n
t
t
es
t
(
F
AS
T
)
a
n
d
t
h
e
m
u
l
ti
-
r
e
s
o
l
u
t
i
o
n
a
n
a
l
y
s
is
(
MR
A
)
i
s
u
s
e
d
f
o
r
c
l
a
s
s
i
f
ic
a
t
i
o
n
.
A
V
G
GN
E
T
1
9
p
r
e
-
t
r
a
i
n
e
d
m
o
d
e
l
i
s
u
s
e
d
f
o
r
d
e
t
e
c
t
i
n
g
g
u
n
a
n
d
k
n
i
f
e
i
n
h
a
n
d
s
o
f
a
p
e
r
s
o
n
[
1
7
]
.
T
h
e
s
e
s
y
s
t
e
m
s
a
r
e
ac
c
u
r
a
t
e
,
b
u
t
t
h
e
h
a
n
d
g
u
n
d
e
t
e
c
t
i
o
n
e
s
p
ec
i
a
l
l
y
w
h
e
n
c
r
i
m
e
s
o
c
c
u
r
n
e
ed
a
f
a
s
t
e
r
a
n
d
m
o
r
e
a
c
c
u
r
a
t
e
s
y
s
t
e
m
.
3.
T
H
E
P
RO
P
O
SE
D
M
O
D
E
L
T
h
e
m
a
i
n
i
d
e
a
o
f
o
b
j
e
c
t
d
e
t
e
c
t
i
o
n
i
s
t
o
r
e
c
o
g
n
i
z
e
t
h
e
o
b
j
e
c
t
i
n
t
h
e
i
n
p
u
t
i
m
a
g
e
a
n
d
f
i
n
d
i
t
s
l
o
c
a
t
i
o
n
[
1
8
]
.
T
h
e
d
e
s
i
g
n
e
d
s
y
s
t
e
m
f
o
c
u
s
e
s
o
n
d
e
t
e
c
t
i
n
g
h
a
n
d
g
u
n
s
i
n
m
i
n
i
m
u
m
t
r
a
i
n
i
n
g
t
i
m
e
w
i
t
h
h
i
g
h
a
c
c
u
r
a
c
y
r
e
s
u
l
t
s
.
A
p
r
e
-
t
r
a
i
n
e
d
m
o
d
e
l
s
u
c
h
a
s
G
o
o
g
l
e
N
e
t
,
V
G
G
N
e
t
-
1
9
a
n
d
M
o
b
i
l
e
N
e
t
h
a
v
e
b
e
e
n
t
r
a
i
n
e
d
w
i
t
h
m
o
r
e
t
h
a
n
m
i
l
l
i
o
n
i
m
a
g
e
s
t
o
m
i
n
i
m
i
z
e
t
h
e
o
b
j
e
c
t
d
e
t
e
c
t
i
o
n
e
r
r
o
r
s
i
n
i
m
a
g
e
s
[
1
9
-
2
1
]
.
B
a
s
e
d
o
n
f
a
s
t
a
n
d
a
c
c
u
r
a
t
e
l
y
t
r
a
i
n
i
n
g
p
r
o
p
e
r
t
i
e
s
,
a
M
o
b
i
l
e
N
e
t
v
3
m
o
d
e
l
i
s
u
s
e
d
i
n
t
h
i
s
m
e
t
h
o
d
t
o
c
l
a
s
s
i
f
y
a
n
d
d
e
t
e
c
t
h
a
n
d
g
u
n
s
a
c
c
u
r
a
t
e
l
y
.
E
x
p
e
r
i
m
e
n
t
a
l
l
y
,
t
h
e
s
y
s
t
e
m
i
s
t
e
s
t
e
d
w
i
t
h
a
n
i
m
a
g
e
d
a
t
a
s
e
t
w
h
i
c
h
c
o
n
t
a
i
n
s
v
a
r
i
o
u
s
i
m
a
g
e
s
o
f
w
e
a
p
o
n
s
a
n
d
d
i
f
f
e
r
e
n
t
p
o
s
i
t
i
o
n
o
f
g
u
n
s
l
i
n
g
e
r
.
T
h
e
b
u
ild
in
g
o
f
m
o
b
ile
m
o
d
e
ls
is
b
ased
o
n
en
h
a
n
cin
g
th
e
ef
f
icien
cy
f
o
r
th
e
b
u
ild
in
g
b
l
o
ck
s
[
2
2
]
.
Mo
b
ileNetV1
[
2
3
]
p
r
esen
ted
d
e
p
t
h
w
i
s
e
s
ep
ar
ab
le
co
n
v
o
lu
ti
o
n
s
(
DSC
)
as
an
ef
f
icien
t
ch
an
g
e
f
o
r
o
th
e
r
C
NN
lay
er
s
.
DSC
is
u
tili
ze
d
to
d
ec
o
m
p
o
s
e
t
h
e
tr
a
d
itio
n
al
c
o
n
v
o
lu
tio
n
in
t
o
d
ep
th
wis
e
co
n
v
o
lu
t
io
n
an
d
p
o
in
twis
e
co
n
v
o
l
u
tio
n
,
in
Mo
b
ileNet
[
1
8
]
.
I
n
D
ep
th
wis
e
co
n
v
o
lu
tio
n
a
p
p
r
o
ac
h
,
a
s
in
g
le
c
o
n
v
o
lu
tio
n
al
f
ilter
is
ap
p
lied
f
o
r
ea
ch
in
p
u
t c
h
a
n
n
el,
wh
er
ea
s
t
h
e
P
o
in
twis
e
co
n
v
o
lu
tio
n
p
er
f
o
r
m
s
a
1
*
1
co
n
v
o
lu
tio
n
to
co
m
b
in
e
th
o
s
e
s
ep
ar
ate
ch
an
n
els
as sh
o
wn
in
Fig
u
r
e
1
.
(
a)
(
b
)
(
c)
Fig
u
r
e
1
.
Dep
t
h
-
wis
e
s
ep
ar
ab
l
e
co
n
v
o
lu
tio
n
DSC
in
Mo
b
ile
Net
;
(
a)
s
tan
d
ar
d
co
n
v
o
lu
tio
n
,
(
b
)
d
ep
th
wis
h
co
n
v
o
lu
tio
n
al
f
ilter
s
,
(
c)
1
x
1
co
n
v
o
lu
ti
o
n
al
f
ilter
s
c
alled
p
o
in
twis
e
in
th
e
co
n
tex
t
o
f
d
e
p
th
s
ep
ar
ab
le
co
n
v
o
l
u
tio
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6
9
3
0
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
,
Vo
l.
18
,
No
.
6
,
Dec
em
b
e
r
2
0
2
0
:
30
26
-
303
2
3028
T
h
e
n
o
v
el
d
esig
n
f
o
r
th
is
m
o
d
el
lead
s
to
r
ed
u
ce
b
o
th
o
f
th
e
co
m
p
lex
ity
o
f
c
o
m
p
u
tatio
n
an
d
th
e
p
ar
am
eter
s
n
u
m
b
er
.
Mo
b
ileNetV2
[
24
]
p
r
esen
te
d
th
e
lin
e
ar
b
o
ttlen
ec
k
an
d
in
v
er
te
d
r
esid
u
al
m
o
d
el
s
o
as
to
g
et
a
m
u
ch
e
f
f
icien
t
s
tr
u
ctu
r
e
f
o
r
th
e
C
NN
lay
er
s
b
y
u
tili
zin
g
th
e
lo
w
r
an
k
n
atu
r
e
o
f
th
e
p
r
o
b
lem
.
T
h
e
ar
ch
itectu
r
e
o
f
Mo
b
ileNe
tV2
is
b
u
ilt
b
ased
on
th
e
s
tr
u
ctu
r
e
of
in
v
e
r
ted
r
esid
u
al
as
s
h
o
wn
in
Fig
u
r
e
2
,
wh
er
e
th
e
i
n
p
u
t
a
n
d
o
u
tp
u
t
c
o
n
n
ec
tio
n
s
o
f
th
e
r
esid
u
al
b
l
o
ck
ar
e
th
in
b
o
ttlen
ec
k
lay
er
s
.
T
h
e
in
ter
m
e
d
iate
ex
p
an
s
io
n
lay
e
r
u
tili
ze
s
th
e
lig
h
tweig
h
t
d
ep
t
h
wis
e
co
n
v
o
l
u
tio
n
s
to
f
ilter
th
e
f
ea
tu
r
es.
Mn
asNet
[
25
]
is
b
u
ilt
b
ased
u
p
o
n
th
e
s
tr
u
ctu
r
e
o
f
a
Mo
b
ileNetV2
m
o
d
el
wh
ich
p
r
esen
t
s
th
e
lig
h
tweig
h
t
atten
tio
n
m
o
d
u
les
b
ased
o
n
s
q
u
ee
ze
an
d
ex
citatio
n
in
to
th
e
b
o
ttlen
ec
k
s
tr
u
ctu
r
e.
T
h
e
d
esg
in
ed
m
o
d
u
le
is
p
lace
d
in
th
e
e
x
p
an
s
io
n
,
f
o
llo
wed
by
th
e
d
ep
th
wis
e
f
ilter
s
to
ap
p
ly
it o
n
th
e
lar
g
est r
ep
r
esen
tati
o
n
as sh
o
wn
o
n
Fig
u
r
e
3
.
Fig
u
r
e
2
.
Mo
b
ileNetV2
lay
er
(
in
v
er
ted
r
esid
u
al
a
n
d
lin
ea
r
b
o
ttlen
ec
k
)
Fig
u
r
e
3
.
Mo
b
ileNetV2
with
s
q
u
ee
ze
-
an
d
-
ex
cite
T
o
b
u
ild
m
o
r
e
ef
f
icin
t
m
o
d
el
,
Mo
b
ileNetV3
is
co
m
b
in
ed
to
im
p
r
o
v
e
d
th
ese
lay
er
s
with
m
o
d
if
ied
s
wis
h
n
o
n
lin
ea
r
ities
.
it
is
r
ep
lace
d
th
e
s
ig
m
o
id
ac
tiv
atio
n
f
u
n
ctio
n
b
y
h
ar
d
s
ig
m
o
id
b
e
ca
u
s
e
its
ea
s
ier
to
ca
lcu
late
an
d
m
o
r
e
ac
cu
r
a
te
.
Mo
b
ileNetV3
h
as
two
m
o
d
els,
wh
ich
ar
e
Mo
b
ileNetV3
-
L
ar
g
e
an
d
Mo
b
ileNetV3
-
Sm
all.
T
h
e
M
o
b
ileNet
V3
-
L
ar
g
e
h
as
a
h
ig
h
er
r
eso
u
r
ce
th
an
t
h
e
Mo
b
ileNetV3
-
Sm
all,
th
e
s
p
ec
if
icatio
n
f
o
r
b
o
t
h
n
etwo
r
k
s
ar
e
s
h
o
wn
in
T
ab
le
s
1
an
d
2
.
T
ab
le
1
.
Sp
ec
if
icatio
n
f
o
r
th
e
m
o
b
ileNet
V3
-
lar
g
e
I
n
p
u
t
O
p
e
r
a
t
o
r
e
x
p
s
i
z
e
#
o
u
t
SE
NL
s
22
4
2
x
3
c
o
n
v
2
d
-
16
-
HS
2
1
1
2
2
x16
b
n
e
c
,
3
x
3
16
16
-
RE
1
1
1
2
2
x
16
b
n
e
c
,
3
x
3
64
24
-
RE
2
56
2
x
24
b
n
e
c
,
3
x
3
72
24
-
RE
1
56
2
x
24
b
n
e
c
,
5
x
5
72
40
✓
RE
2
28
2
x
40
b
n
e
c
,
5
x
5
1
2
0
40
✓
RE
1
28
2
x
40
b
n
e
c
,
5
x
5
1
2
0
40
✓
RE
1
28
2
x
40
b
n
e
c
,
3
x
3
2
4
0
80
-
HS
2
14
2
x
80
b
n
e
c
,
3
x
3
2
0
0
80
-
HS
1
14
2
x
80
b
n
e
c
,
3
x
3
1
8
4
80
-
HS
1
14
2
x
80
b
n
e
c
,
3
x
3
1
8
4
80
-
HS
1
14
2
x
80
b
n
e
c
,
3
x
3
4
8
0
1
1
2
✓
HS
1
14
2
x
1
1
2
b
n
e
c
,
3
x
3
6
7
2
1
1
2
✓
HS
1
14
2
x
1
1
2
b
n
e
c
,
5
x
5
6
7
2
1
6
0
✓
HS
1
14
2
x
1
1
2
b
n
e
c
,
5
x
5
6
7
2
1
6
0
✓
HS
2
7
2
x
1
6
0
b
n
e
c
,
5
x
5
9
6
0
1
6
0
✓
HS
1
7
2
x
1
6
0
c
o
n
v
2
d
,
1
x
1
-
9
6
0
-
HS
1
7
2
x
9
6
0
P
o
o
l
,
7
x
7
-
-
-
HS
-
1
2
x
9
6
0
c
o
n
v
2
d
1
x
1
,
n
b
n
-
1
2
8
0
-
HS
1
1
2
x
1
2
8
0
c
o
n
v
2
d
1
x
1
,
n
b
n
-
k
-
-
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
T
h
e
d
etec
tio
n
o
f
h
a
n
d
g
u
n
s
fr
o
m
live
-
vid
eo
in
r
ea
l time
b
a
s
e
d
o
n
d
ee
p
lea
r
n
in
g
(
Mu
h
a
mme
d
Gh
a
z
a
l
)
3029
T
ab
le
2
.
Sp
ec
if
icatio
n
f
o
r
Mo
b
ileNetV3
-
s
m
all
I
n
p
u
t
O
p
e
r
a
t
o
r
e
x
p
s
i
z
e
#
o
u
t
SE
NL
2
2
4
2
x
3
c
o
n
v
2
d
,
3
x
3
-
16
-
HS
1
1
2
2
x24
b
n
e
c
,
3
x
3
16
16
✓
RE
56
2
x
24
b
n
e
c
,
3
x
3
72
24
-
RE
28
2
x
24
b
n
e
c
,
3
x
3
88
24
-
RE
28
2
x
40
b
n
e
c
,
5
x
5
96
40
✓
HS
14
2
x
40
b
n
e
c
,
5
x
5
2
4
0
40
✓
HS
14
2
x
40
b
n
e
c
,
5
x
5
2
4
0
40
✓
HS
14
2
x
40
b
n
e
c
,
5
x
5
1
2
0
48
✓
HS
14
2
x
48
b
n
e
c
,
5
x
5
1
4
4
48
✓
HS
14
2
x
96
b
n
e
c
,
5
x
5
2
8
8
96
✓
HS
7
2
x
96
b
n
e
c
,
5
x
5
5
7
6
96
✓
HS
7
2
x
96
b
n
e
c
,
5
x
5
5
7
6
96
✓
HS
7
2
x
96
c
o
n
v
2
d
,
1
x
1
-
5
7
6
✓
HS
7
2
x
5
7
6
P
o
o
l
,
7
x
7
-
-
-
HS
1
2
x
5
7
6
c
o
n
v
2
d
,
1
x
1
-
1
2
8
0
-
HS
1
2
x
1
2
8
0
c
o
n
v
2
d
,
1
x
1
-
k
-
HS
4.
E
XP
E
R
I
M
E
N
T
A
L
M
O
D
E
L
T
RAIN
I
NG
T
h
e
ex
p
er
im
e
n
ts
h
av
e
b
ee
n
ac
co
m
p
lis
h
ed
b
y
u
s
in
g
a
wo
r
k
s
tatio
n
eq
u
ip
p
ed
with
an
I
n
tel
C
o
r
e
I
7
-
7
7
0
0
C
PU
@
3
.
6
0
GHZ
,
NVI
DI
A
Gr
ap
h
ics
Pro
ce
s
s
in
g
Un
it
(
GPU)
o
f
3
8
4
0
C
UDA
co
r
es
an
d
1
6
GB
DDR4
R
AM
2
4
0
0
MH
z,
with
Ub
u
n
tu
1
6
.
0
4
o
p
e
r
atin
g
s
y
s
tem
.
T
o
e
v
alu
a
te
th
e
p
r
o
p
o
s
ed
m
o
d
el,
th
r
ee
d
atasets
ar
e
u
s
ed
f
o
r
th
e
lear
n
in
g
p
h
ase:
-
T
h
e
h
a
n
d
g
u
n
d
a
t
a
s
e
t
o
f
s
l
i
d
i
n
g
w
i
n
d
o
w
a
p
p
r
o
a
c
h
,
w
h
i
c
h
c
o
n
s
i
s
t
s
o
f
1
0
2
c
l
a
s
s
e
s
w
i
t
h
a
r
o
u
n
d
9
2
6
1
i
m
a
g
e
s
[
1
1
]
.
-
T
h
e
h
an
d
g
u
n
d
ataset
o
f
r
eg
i
o
n
p
r
o
p
o
s
als ap
p
r
o
ac
h
,
wh
ic
h
co
n
s
is
ts
o
f
6
0
8
im
a
g
es a
n
d
i
n
clu
d
e
s
3
0
4
im
ag
es
o
f
h
an
d
g
u
n
s
[
1
1
]
.
-
W
ea
p
o
n
d
ataset,
wh
ich
co
n
tain
s
3
0
0
0
im
ag
es o
f
g
u
n
s
tak
e
n
in
a
v
ar
iety
o
f
co
n
tex
ts
[
1
2
]
.
Fig
u
r
e
4
s
h
o
ws
s
o
m
e
s
am
p
le
s
o
f
h
an
d
g
u
n
d
ataset
.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
tr
ain
e
d
with
c
af
f
e
d
ee
p
lear
n
in
g
lib
r
ar
y
to
ap
p
r
o
ac
h
h
an
d
g
u
n
d
etec
tio
n
in
liv
e
v
id
e
o
s
with
n
ea
r
to
r
ea
l
tim
e
.
Fig
u
r
e
5
illu
s
tr
ates
th
e
h
an
d
g
u
n
s
d
etec
tio
n
s
y
s
tem
s
tr
u
ctu
r
e.
At
f
ir
s
t,
th
e
p
r
o
g
r
a
m
in
itializes
th
e
tr
ain
ed
lis
t
o
f
class
lab
els
b
y
Mo
b
ileNetV3
-
SS
DL
ite
to
d
e
tect
h
an
d
g
u
n
s
.
Af
te
r
th
at,
a
s
et
o
f
b
o
u
n
d
in
g
b
o
x
c
o
lo
r
s
ar
e
g
en
er
ate
d
f
o
r
ea
c
h
class
b
ef
o
r
e
lo
ad
in
g
th
e
p
r
o
p
o
s
ed
m
o
d
el.
T
h
en
,
b
e
g
in
in
g
t
o
in
p
u
t
im
ag
es
an
d
r
esize
th
em
to
t
h
e
s
ize
o
f
3
0
0
x
3
0
0
p
ix
els
,
an
d
th
en
c
o
n
v
er
t
t
h
e
m
to
b
lo
b
s
.
Fin
aly
,
p
ass
in
g
th
e
b
l
o
b
s
th
r
o
u
g
h
th
e
C
NN
n
etwo
r
k
to
o
b
tain
d
etec
tio
n
an
d
p
r
e
d
icatio
n
s
o
f
h
an
d
g
u
n
s
.
E
ac
h
tim
e
th
e
h
a
n
d
g
u
n
d
e
t
ec
to
r
f
in
d
s
a
wea
p
o
n
,
th
e
C
NN
m
o
d
el
r
etu
r
n
s
b
o
u
n
d
ar
y
b
o
x
ar
o
u
n
d
th
e
o
b
ject
(
h
an
d
g
u
n
o
r
an
y
o
th
er
t
y
p
es
o
f
d
ef
in
ed
wea
p
o
n
s
in
th
e
d
atasets
)
,
with
d
eter
m
in
ed
ce
n
tr
o
id
,
th
e
s
y
s
tem
s
en
d
s
an
aler
t m
ess
ag
e
to
th
e
s
u
p
er
v
is
o
r
.
Fig
u
r
e
4
.
S
am
p
les
o
f
wea
p
o
n
d
at
aset
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6
9
3
0
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
,
Vo
l.
18
,
No
.
6
,
Dec
em
b
e
r
2
0
2
0
:
30
26
-
303
2
3030
Fig
u
r
e
5
.
Han
d
g
u
n
d
etec
tio
n
s
y
s
tem
s
tr
u
ctu
r
e
5.
RE
SU
L
T
S
T
h
e
f
i
n
d
i
n
g
s
o
f
t
h
i
s
m
e
t
h
o
d
s
u
g
g
e
s
t
t
h
a
t
M
o
b
il
e
N
et
V
3
i
s
s
u
i
ta
b
l
e
f
o
r
h
a
n
d
g
u
n
d
e
t
e
c
t
i
o
n
a
s
i
t
d
e
l
i
v
e
r
s
a
p
e
r
f
e
c
t
c
o
r
r
e
s
p
o
n
d
e
n
c
e
b
et
w
ee
n
p
r
e
d
i
c
t
i
o
n
s
p
e
e
d
a
n
d
a
c
c
u
r
a
c
y
.
T
h
e
p
r
o
p
o
s
e
d
m
o
d
e
l
a
c
h
i
e
v
e
d
9
6
%
t
r
a
i
n
i
n
g
a
c
c
u
r
a
c
y
r
a
t
e
w
h
i
le
t
h
e
r
e
s
u
lt
o
f
p
r
e
v
i
o
u
s
n
e
tw
o
r
k
s
:
M
o
b
i
l
eN
et
V
2
-
S
SD
a
n
d
G
o
o
g
l
e
N
et
i
n
c
e
p
t
i
o
n
w
e
r
e
8
9
%
a
n
d
8
7
%
r
e
s
p
e
c
ti
v
e
l
y
.
T
h
e
s
p
e
e
d
ac
h
i
e
v
e
d
b
y
t
h
e
e
x
p
e
r
i
m
e
n
t
u
s
i
n
g
M
o
b
i
l
e
N
e
tV
3
-
S
SD
L
i
te
i
s
∼
20
–
2
3
F
P
S
,
w
h
e
r
e
as
M
o
b
i
l
e
N
et
V
2
-
SS
D
h
as
a
c
h
i
ev
e
d
∼
15
–
1
8
F
PS
.
I
n
t
e
r
m
s
o
f
p
r
e
d
i
c
t
i
o
n
,
M
o
b
il
e
N
e
tV
3
-
SS
D
L
i
t
e
h
as
t
h
e
b
e
s
t
p
e
r
f
o
r
m
a
n
c
e
0
.
9
4
2
f
o
r
f
e
a
t
u
r
e
e
x
t
r
a
c
t
e
r
w
h
i
l
e
M
o
b
i
l
eN
e
tV
2
-
S
S
D
h
a
s
0
.
8
1
6
.
T
h
e
e
x
p
e
r
i
m
e
n
t
s
h
a
v
e
b
e
e
n
p
e
r
f
o
r
m
e
d
i
n
a
r
e
al
-
t
i
m
e
d
e
t
e
cti
o
n
i
n
li
v
e
v
i
d
e
o
s
a
n
d
i
m
a
g
e
s
as
s
h
o
w
n
i
n
F
i
g
u
r
e
6
.
T
h
e
s
y
s
t
em
i
s
p
r
o
v
i
d
e
d
w
it
h
S
M
S
f
ea
t
u
r
e
,
w
h
i
c
h
al
l
o
w
t
h
e
s
y
s
t
e
m
t
o
s
e
n
d
s
a
n
a
le
r
t
m
es
s
a
g
e
t
o
t
h
e
s
u
p
e
r
v
is
o
r
w
h
e
n
e
v
e
r
a
h
a
n
d
g
u
n
i
s
d
e
t
e
c
t
e
d
.
Fig
u
r
e
6
.
E
x
am
p
le
o
f
h
a
n
d
g
u
n
d
etec
tio
n
s
y
s
tem
6.
CO
NCLU
SI
O
N
T
h
e
p
r
o
p
o
s
ed
Han
d
g
u
n
Dete
ctio
n
Sy
s
tem
is
r
ea
lly
an
ap
p
licab
le
s
y
s
tem
wh
ich
h
elp
s
to
d
etec
t c
r
im
es
an
d
g
u
n
s
lin
g
er
.
T
h
e
u
s
e
o
f
M
o
b
ileNetV3
p
r
e
-
tr
ain
ed
m
o
d
el
is
m
u
ch
ef
f
icien
t
b
ec
a
u
s
e
it
i
s
f
aster
th
an
o
th
e
r
m
o
d
els,
s
u
ch
as
Mo
b
ileNetV2
an
d
Go
o
g
leNe
t
.
Mo
r
eo
v
er
,
th
e
s
ec
u
r
ity
alar
m
p
r
o
p
er
ty
in
cr
ea
s
e
s
s
y
s
tem
ef
f
ec
tiv
en
ess
.
T
h
e
p
r
o
p
o
s
ed
s
y
s
tem
ca
n
b
e
u
s
ed
in
v
ar
io
u
s
ap
p
licatio
n
s
,
s
u
ch
as,
r
ea
l
tim
e
d
etec
tio
n
o
f
g
u
n
s
in
s
u
p
er
m
ar
k
ets
wh
ic
h
is
m
o
n
ito
r
ed
b
y
ca
m
er
as
,
a
n
d
also
ch
ec
k
in
g
w
h
eth
er
th
e
u
p
lo
a
d
ed
v
id
eo
s
to
y
o
u
tu
b
e
a
n
d
o
th
er
s
o
cial
m
ed
ia
n
etwo
r
k
s
in
clu
d
e
s
ce
n
es with
g
u
n
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
T
h
e
d
etec
tio
n
o
f
h
a
n
d
g
u
n
s
fr
o
m
live
-
vid
eo
in
r
ea
l time
b
a
s
e
d
o
n
d
ee
p
lea
r
n
in
g
(
Mu
h
a
mme
d
Gh
a
z
a
l
)
3031
ACK
NO
WL
E
DG
E
M
E
NT
S
T
h
is
r
esear
ch
is
f
u
n
d
ed
b
y
co
o
p
er
atio
n
b
etwe
en
Dee
p
L
ea
r
n
in
g
team
in
No
r
th
er
n
T
ec
h
n
ical
Un
iv
er
s
ity
NT
U
in
I
r
aq
.
W
eb
s
ite: Fo
r
No
r
th
er
n
T
ec
h
n
ical
Un
iv
er
s
ity
h
t
tp
s
://www.
n
tu
.
ed
u
.
iq
RE
F
E
R
E
NC
E
S
[1
]
S
tu
ru
p
J
.
,
e
t
a
l
.
,
“
In
c
re
a
se
d
g
u
n
v
io
len
c
e
a
m
o
n
g
y
o
u
n
g
m
a
les
in
S
we
d
e
n
:
a
d
e
sc
rip
ti
v
e
n
a
ti
o
n
a
l
su
rv
e
y
a
n
d
in
tern
a
ti
o
n
a
l
c
o
m
p
a
,”
Eu
ro
p
e
a
n
J
o
u
rn
a
l
o
n
Crimin
a
l
Po
li
c
y
a
n
d
Res
e
a
rc
h
,
v
o
l.
2
5
,
no.
4
,
p
p
.
365
-
3
7
8
,
M
a
y
2
0
1
8
.
[2
]
Un
it
e
d
Na
ti
o
n
s Office
o
n
Dru
g
s
a
n
d
Crime
(UN
OD
C),
“
G
lo
b
a
l
stu
d
y
o
n
h
o
m
icid
e
2
0
1
9
,
”
Da
ta:
UN
OD
C
Ho
m
icid
e
S
tatisti
c
s
,
2019.
[3
]
Ch
e
n
W
.
,
e
t
a
l
.
,
“
All
y
o
u
n
e
e
d
is
a
fe
w
sh
ift
s:
De
sig
n
in
g
e
fficie
n
t
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
s
fo
r
ima
g
e
c
las
sifica
ti
o
n
”
.
Pro
c
e
e
d
in
g
s o
f
th
e
IEE
E
Co
n
fer
e
n
c
e
o
n
Co
mp
u
ter
Vi
sio
n
a
n
d
Pa
tt
e
rn
Rec
o
g
n
it
io
n
,
Ju
n
e
2019.
[4
]
Li
u
L
.
,
e
t
a
l
.
,
“
De
e
p
lea
rn
in
g
fo
r
g
e
n
e
ric
o
b
jec
t
d
e
tec
ti
o
n
:
A
su
rv
e
y
”
.
In
ter
n
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
Co
mp
u
ter
Vi
sio
n
,
v
o
l.
1
2
8
,
p
p
.
261
–
318
,
S
e
p
tem
b
e
r
2
0
1
8
.
[5
]
Zo
p
h
B,
e
t
a
l.
,
“
Lea
rn
in
g
tran
sfe
ra
b
le
a
rc
h
it
e
c
tu
re
s
fo
r
sc
a
lab
le
ima
g
e
re
c
o
g
n
it
io
n
”
.
Pro
c
e
e
d
in
g
s
o
f
th
e
IEE
E
c
o
n
fer
e
n
c
e
o
n
c
o
mp
u
ter
v
isio
n
a
n
d
p
a
tt
e
rn
re
c
o
g
n
it
io
n
,
Ju
n
e
2018.
[6
]
Li
Z
.
,
G
a
v
ril
y
u
k
K
.
,
G
a
v
v
e
s
E
.
,
Ja
in
M
.
,
S
n
o
e
k
C
.
G.
,
“
Vid
e
o
LS
TM
c
o
n
v
o
lv
e
s,
a
tt
e
n
d
s
a
n
d
flo
ws
fo
r
a
c
ti
o
n
re
c
o
g
n
it
io
n
,”
Co
mp
u
ter
Vi
sio
n
a
n
d
Ima
g
e
Un
d
e
rs
ta
n
d
in
g
,
v
o
l.
1
6
6
,
p
p
.
4
1
-
5
0
,
Ja
n
u
a
ry
2018
.
[7
]
M
a
rtí
n
e
z
F
.
,
He
rn
á
n
d
e
C
.
,
M
a
rtí
n
e
z
F
.
E.
,
“
Ev
a
lu
a
ti
o
n
o
f
d
e
e
p
n
e
u
ra
l
n
e
two
rk
a
rc
h
it
e
c
tu
re
s
in
th
e
id
e
n
ti
fica
ti
o
n
o
f
b
o
n
e
fissu
re
s
,”
T
EL
KOM
NIKA
Te
lec
o
mm
u
n
ica
ti
o
n
Co
mp
u
ti
n
g
El
e
c
tro
n
ics
a
n
d
Co
n
tro
l
,
v
o
l.
1
8
,
n
o
.
2
,
p
p
.
807
-
8
1
4
,
Ap
ril
2
0
2
0
.
[8
]
He
re
d
ia
A,
Ba
rro
s
-
G
a
v
il
a
n
e
s
G
,
e
d
it
o
rs.
“
Vid
e
o
p
ro
c
e
ss
in
g
in
sid
e
e
m
b
e
d
d
e
d
d
e
v
ice
s
u
sin
g
ss
d
-
m
o
b
il
e
n
e
t
to
c
o
u
n
t
m
o
b
il
it
y
a
c
to
rs
,”
2
0
1
9
IEE
E
Co
lo
mb
ia
n
Co
n
fer
e
n
c
e
o
n
Ap
p
li
c
a
ti
o
n
s
in
Co
mp
u
ta
ti
o
n
a
l
In
telli
g
e
n
c
e
(Co
lCA
CI)
,
Ju
n
e
2
0
1
9
.
[9
]
Ho
wa
rd
A
.
,
e
t
a
l
.
,
“
S
e
a
rc
h
in
g
fo
r
m
o
b
il
e
n
e
t
v
3
”
,
Pr
o
c
e
e
d
in
g
s
o
f
t
h
e
IEE
E
In
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
Co
mp
u
ter
Vi
sio
n
,
No
v
e
m
b
e
r
2
0
1
9
.
[1
0
]
S
imo
n
y
a
n
K,
Zi
ss
e
rm
a
n
A.
“
Ve
ry
d
e
e
p
c
o
n
v
o
lu
ti
o
n
a
l
n
e
two
rk
s
fo
r
larg
e
-
sc
a
le
ima
g
e
re
c
o
g
n
it
io
n
”
.
a
rXiv
p
re
p
rin
t
a
rXiv
,
S
e
p
tem
b
e
r
2
0
1
4
.
[1
1
]
El
m
ir
Y
.
,
Lao
u
a
r
S
.
A
.
,
Ha
m
d
a
o
u
i
L
.
,
“
De
e
p
lea
rn
in
g
fo
r
a
u
to
m
a
ti
c
d
e
tec
ti
o
n
o
f
h
a
n
d
g
u
n
s
in
v
id
e
o
se
q
u
e
n
c
e
s
,
”
3
rd
e
d
it
io
n
o
f
th
e
Na
ti
o
n
a
l
S
tu
d
y
Da
y
o
n
Res
e
a
rc
h
o
n
Co
mp
u
ter
S
c
ien
c
e
s
(J
ER
I
2
0
1
9
),
S
a
id
a
,
Alg
e
ria,
v
o
l.
2
3
5
1
,
p
p
.
1
-
10,
Ap
ril
2
7
,
2
0
1
9
.
[On
li
n
e
].
Av
a
il
a
b
le:
h
tt
p
:/
/ce
u
r
-
ws
.
o
rg
/Vo
l
-
2
3
5
1
/p
a
p
e
r_
6
9
.
p
d
f
.
[1
2
]
Olm
o
s
R
.
,
Tab
ik
S
.
,
He
rre
ra
F
.
,
“
Au
to
m
a
ti
c
h
a
n
d
g
u
n
d
e
tec
ti
o
n
a
larm
in
v
id
e
o
s
u
sin
g
d
e
e
p
lea
rn
in
g
,
”
Ne
u
ro
c
o
mp
u
ti
n
g
,
v
o
l.
2
7
5
,
p
p
.
6
6
-
7
2
,
F
e
b
ru
a
ry
2018
.
[1
3
]
F
li
tt
o
n
G
.
,
Bre
c
k
o
n
T
.
P
.
,
M
e
g
h
e
rb
i
N.
,
“
A
c
o
m
p
a
riso
n
o
f
3
D
in
tere
st
p
o
in
t
d
e
sc
rip
to
rs
with
a
p
p
li
c
a
ti
o
n
to
a
irp
o
r
t
b
a
g
g
a
g
e
o
b
jec
t
d
e
tec
ti
o
n
in
c
o
m
p
lex
CT
ima
g
e
ry
,”
Pa
tt
e
rn
Rec
o
g
n
it
io
n
,
v
o
l.
4
6
,
n
o
.
9
,
p
p
.
2
4
2
0
-
2
4
3
6
,
S
e
p
tem
b
e
r
2
0
1
3
.
[1
4
]
Ti
wa
ri
R
.
K
.
,
Ve
rm
a
G
.
K.
,
“
A
c
o
m
p
u
ter
v
isio
n
b
a
se
d
fra
m
e
wo
rk
fo
r
v
isu
a
l
g
u
n
d
e
tec
ti
o
n
u
sin
g
h
a
rris
in
tere
st
p
o
in
t
d
e
tec
to
r
,
”
Pro
c
e
d
ia
Co
mp
u
ter
S
c
ien
c
e
,
v
o
l.
5
4
,
p
p
.
7
0
3
-
712,
2015
.
[1
5
]
Ti
wa
ri
R
.
K
.
,
Ve
rm
a
G
.
K
.
,
“
A
c
o
m
p
u
ter
v
isio
n
b
a
se
d
fra
m
e
wo
rk
fo
r
v
isu
a
l
g
u
n
d
e
tec
ti
o
n
u
sin
g
S
URF”.
2
0
1
5
In
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
El
e
c
trica
l,
El
e
c
tro
n
ics
,
S
ig
n
a
ls,
Co
mm
u
n
ica
ti
o
n
a
n
d
Op
ti
miza
ti
o
n
(EE
S
CO)
,
Ja
n
u
a
ry
2015.
[1
6
]
Bu
c
k
c
h
a
sh
H
.
,
e
t
a
l
.
,
“
A
ro
b
u
st o
b
jec
t
d
e
tec
to
r:
a
p
p
li
c
a
ti
o
n
to
d
e
tec
ti
o
n
o
f
v
isu
a
l
k
n
iv
e
s
,”
2
0
1
7
IEE
E
In
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
M
u
lt
ime
d
ia
&
Exp
o
W
o
rk
sh
o
p
s (ICM
EW
)
,
Ju
ly
2017
.
[1
7
]
Na
v
a
lg
u
n
d
U
.
V
.
,
P
riy
a
d
h
a
rsh
in
i
K
.
,
“
Crime
in
ten
ti
o
n
d
e
tec
ti
o
n
sy
ste
m
u
sin
g
d
e
e
p
lea
rn
in
g
,”
2
0
1
8
In
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
Circ
u
it
s a
n
d
S
y
ste
ms
in
Dig
it
a
l
En
ter
p
rise
T
e
c
h
n
o
lo
g
y
(ICCS
DET
)
,
De
c
e
m
b
e
r
2
0
1
8
.
[1
8
]
F
e
n
g
X
.
,
e
t
a
l
.
,
“
Co
m
p
u
ter
v
isio
n
a
lg
o
rit
h
m
s
a
n
d
h
a
rd
wa
re
imp
lem
e
n
tatio
n
s:
A
su
rv
e
y
,”
In
teg
ra
ti
o
n
,
v
o
l.
6
9
,
p
p
.
3
0
1
9
-
3
2
0
,
No
v
e
m
b
e
r
2
0
1
9
.
[1
9
]
M
a
su
d
M
.
,
e
t
a
l
.
,
“
De
e
p
lea
rn
in
g
-
b
a
se
d
in
telli
g
e
n
t
fa
c
e
re
c
o
g
n
it
io
n
in
Io
T
-
c
lo
u
d
e
n
v
iro
n
m
e
n
t
,”
Co
mp
u
te
r
Co
mm
u
n
ica
ti
o
n
s
,
v
o
l.
1
5
2
,
p
p
.
2
1
5
-
2
2
2
,
F
e
b
ru
a
ry
2020
.
[2
0
]
Ku
rd
th
o
n
g
m
e
e
W.
,
“
A co
m
p
a
ra
ti
v
e
stu
d
y
o
f
th
e
e
ffe
c
ti
v
e
n
e
ss
o
f
u
sin
g
p
o
p
u
lar
DN
N o
b
jec
t
d
e
tec
ti
o
n
a
lg
o
rit
h
m
s
fo
r
p
it
h
d
e
tec
ti
o
n
in
c
ro
ss
-
se
c
ti
o
n
a
l
ima
g
e
s o
f
p
a
ra
wo
o
d
,
”
He
li
y
o
n
,
v
o
l.
6
,
n
o
.
2
,
F
e
b
ru
a
ry
2
0
2
0
.
[2
1
]
Attam
imi
M
.
,
M
a
rd
iy
a
n
to
R
.
,
Irfa
n
sy
a
h
A.
,
“
In
c
li
n
e
d
ima
g
e
re
c
o
g
n
i
ti
o
n
fo
r
a
e
rial
m
a
p
p
in
g
u
sin
g
d
e
e
p
lea
rn
in
g
a
n
d
tree
b
a
se
d
m
o
d
e
ls
,”
Te
lec
o
mm
u
n
ica
ti
o
n
,
Co
mp
u
ti
n
g
,
El
e
c
tro
n
ics
a
n
d
Co
n
tro
l
,
v
o
l.
1
6
,
n
o
.
6
,
p
p
.
3
0
3
4
-
3
0
4
4
,
De
c
e
m
b
e
r
2
0
1
8
.
[2
2
]
S
h
a
d
e
e
d
G
.
A
.
,
Taw
fe
e
q
M
.
A
.
,
M
a
h
m
o
u
d
S
.
M.
,
“
De
e
p
lea
rn
in
g
m
o
d
e
l
fo
r
th
o
ra
x
d
ise
a
se
s
d
e
tec
ti
o
n
”
.
Te
lec
o
mm
u
n
ica
ti
o
n
,
Co
mp
u
ti
n
g
,
El
e
c
tro
n
ics
a
n
d
Co
n
tro
l
,
v
o
l.
1
8
,
n
o
.
1
,
p
p
.
4
4
1
-
4
4
9
,
F
e
b
ru
a
ry
2020
.
[2
3
]
Ho
wa
rd
AG
,
e
t
a
l
.
,
“
M
o
b
il
e
n
e
ts:
Eff
icie
n
t
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
s
fo
r
m
o
b
il
e
v
isio
n
a
p
p
li
c
a
ti
o
n
s
,
”
a
rXiv
p
re
p
rin
t
a
rXiv
,
Ap
ril
2017.
[2
4
]
S
a
n
d
ler
M
.
,
e
t
a
l
.
,
“
M
o
b
il
e
n
e
tv
2
:
in
v
e
rted
re
sid
u
a
ls an
d
li
n
e
a
r
b
o
tt
len
e
c
k
s
,
”
Pro
c
e
e
d
in
g
s o
f
th
e
IEE
E
c
o
n
fer
e
n
c
e
o
n
c
o
mp
u
ter
v
isio
n
a
n
d
p
a
tt
e
rn
re
c
o
g
n
it
io
n
,
Ju
n
e
2
0
1
8
.
[2
5
]
Tan
M
.
,
e
t
a
l
.
,
“
M
n
a
sn
e
t:
P
latfo
rm
-
a
wa
re
n
e
u
ra
l
a
rc
h
it
e
c
tu
re
se
a
rc
h
fo
r
m
o
b
il
e
,”
Pro
c
e
e
d
in
g
s
o
f
th
e
IEE
E
Co
n
fer
e
n
c
e
o
n
Co
mp
u
ter
Vi
sio
n
a
n
d
Pa
tt
e
rn
Rec
o
g
n
it
io
n
,
Ju
n
e
2019.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6
9
3
0
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
,
Vo
l.
18
,
No
.
6
,
Dec
em
b
e
r
2
0
2
0
:
30
26
-
303
2
3032
B
I
O
G
RAP
H
I
E
S
O
F
AUTH
O
RS
Mo
h
a
m
m
e
d
Ta
l
a
l
G
h
a
z
a
l
o
b
tain
e
d
h
is M
.
S
c
.
d
e
g
re
e
fr
o
m
Co
m
p
u
ter E
n
g
in
e
e
ri
n
g
Tec
h
n
o
lo
g
y
,
No
rth
e
rn
Tec
h
n
ica
l
Un
iv
e
rsity
,
M
o
su
l
,
Ira
q
i
n
2
0
1
6
.
His M
.
S
c
.
th
e
sis e
n
ti
tl
e
d
:
“
W
h
e
e
lch
a
ir
Ro
b
o
t
Co
n
tr
o
l
Us
in
g
EOG
sig
n
a
l
s”
.
H
is
re
se
a
rc
h
in
tere
sts in
c
lu
d
e
t
h
e
d
e
sig
n
o
f
fa
c
e
re
c
o
g
n
it
i
o
n
a
lg
o
rit
h
m
s,
d
e
e
p
lea
rn
in
g
CNN
s
m
o
d
e
ls an
d
o
b
jec
t
d
e
tec
ti
o
n
.
Na
jwa
n
Zu
h
a
ir
Wa
isi
o
b
tain
e
d
h
e
r
M
.
S
c
.
d
e
g
re
e
fro
m
Co
m
p
u
ter
S
c
ien
c
e
De
p
a
rtme
n
t,
M
o
u
s
l
Un
iv
e
rsity
,
M
o
s
u
l,
Ira
q
in
2
0
1
4
.
He
r
M
.
S
c
.
th
e
sis
e
n
ti
tl
e
d
:
“
De
sig
n
a
n
d
Im
p
lem
e
n
tatio
n
o
f
Cli
e
n
t
Ho
n
e
y
p
o
t”
a
n
d
h
e
r
c
u
rre
n
t
re
se
a
rc
h
in
tere
sts
in
c
l
u
d
e
t
h
e
d
e
e
p
lea
rn
in
g
CNN
s
m
o
d
e
ls
a
n
d
o
b
jec
t
d
e
tec
ti
o
n
.
Na
wa
l
Yo
u
n
is
Abd
u
ll
a
h
o
b
tain
e
d
h
e
r
M
.
S
c
.
d
e
g
re
e
fro
m
C
o
m
p
u
ter
En
g
in
e
e
rin
g
Tec
h
n
o
l
o
g
y
,
No
rth
e
rn
Tec
h
n
ica
l
Un
i
v
e
rsity
,
M
o
su
l
,
Ira
q
i
n
2
0
1
4
.
He
r
M
.
S
c
.
th
e
sis
e
n
ti
tl
e
d
:
“
F
P
G
A
Ba
se
d
Vid
e
o
S
c
e
n
e
Bo
u
n
d
a
ries
De
tec
t
io
n
Us
in
g
E
n
h
a
n
c
e
d
S
o
b
e
l
F
il
t
e
r”
a
n
d
h
e
r
c
u
rre
n
t
re
se
a
rc
h
in
tere
sts in
c
lu
d
e
th
e
d
e
e
p
lea
rn
i
n
g
CNN
s m
o
d
e
ls an
d
o
b
jec
t
d
e
tec
t
io
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.